Thanks, this is a good improvement and the lsfd side looks solid.

One note up front so the request below doesn't read as criticism: the non-blocking statx logic you extracted is my own old libmount code, and you moved it faithfully -- nothing wrong on your side. Seeing it lifted out of libmount context is what made me realise we can do better than the original.

So I'd like to split this. The first two commits are generic lib/fileutils.c work that libmount depends on too, and I'd rather have that reviewed and merged on its own. I'll prepare the statx/fileutils part as a separate PR. Once it's in, please rebase this one on top and drop the first two commits, so #4647 keeps only the lsfd bits.

API I'm going to implement

#if defined(HAVE_STATX) && defined(HAVE_STRUCT_STATX)
# define UL_STATX_ESSENTIAL	(STATX_TYPE | STATX_MODE | STATX_INO)
# define UL_STATX_BASIC		(STATX_BASIC_STATS)

extern void ul_statx_to_stat(const struct statx *stx, struct stat *st);
#else
# define UL_STATX_ESSENTIAL	0
# define UL_STATX_BASIC		0
#endif

extern int ul_safe_stat(const char *target, struct stat *st, int nofollow,
			unsigned int mask);

Three differences from the current version:

• No all_basic boolean. A plain statx mask is more flexible and self-documenting -- callers can pass UL_STATX_BASIC, or compose, e.g. UL_STATX_ESSENTIAL | STATX_ATIME. ul_safe_stat() always ORs UL_STATX_ESSENTIAL into @mask, so 0 is a valid request for the minimum.
• ul_statx_to_stat() honours stx_mask. statx(2) is not obliged to return everything we ask for, and with AT_STATX_DONT_SYNC that's exactly the case we care about. It copies only what the kernel advertises and leaves the rest zeroed. (The original libmount code ignored this -- my fault, not yours.)
• ul_safe_stat() fails with EOPNOTSUPP when the essential attributes are missing from stx_mask, so callers can fall back to stat().

libmount will then call ul_safe_stat(target, st, nofollow, 0) and lsfd ul_safe_stat(path, sb, 0, UL_STATX_BASIC).

Notes on the lsfd part

Nothing blocking, CI is all green.

1. lsfd-cmd/util.c -- the wrappers decide once at startup (probe of /proc/self) and never fall back per call. If statx works on procfs but a particular path or filesystem rejects it, lsfd degrades to an error row where plain stat() would have worked. For -b that's the intended contract; for the default mode a fallback would be nicer, and the EOPNOTSUPP above makes it easy.

2. parse_maps_line() -- when map_files fails and the /proc/PID/maps fallback also fails (stat error, or st_ino/st_dev mismatch), the reported error is always the map_files one; the fallback's own reason is discarded. Also stat_err = errno / readlink_err = errno are a bit fragile: ul_path_readlinkf() and ul_path_statx() return -errno on some paths (e.g. ul_path_get_dirfd() failure) without setting the global errno, so a stale value can leak through. The pattern predates this PR, but there are now more places where it matters.

3. lsfd.1.adoc -- EXIT STATUS sits after the EXAMPLES sections; our convention (findmnt.8, lsns.8) is right after OPTIONS, before ENVIRONMENT. Also missing the blank line after the heading that the other sections have.

4. lsfd.1.adoc, -b -- worth spelling out what the user actually gets when /proc/PID/map_files is inaccessible (CONFIG_CHECKPOINT_RESTORE=n, or insufficient ptrace permission): every file-backed mapping becomes a stat/readlink error row.

5. lsfd_path_statf() -- re-implements ul_path_mkpath() with its own vsnprintf; ul_path_vstatxf() already exists. It also mixes return conventions: -ENAMETOOLONG on the format error but -1 + errno otherwise.

6. usage() -- the new line prints 93 columns, the longest existing one is 81. Maybe exit immediately if non-blocking stat is unavailable.

7. tests/ts/lsfd/mkfds-fuse-hungfs -- both subtests share $MNTDIR. If the fd subtest leaves a hung FUSE mount behind (mkfds timeout, so rmdir fails), the mem subtest's mkdir -p "$MNTDIR" stats the hung mountpoint and can block the whole run. A per-subtest directory, or umount -l in the cleanup, would make it self-healing.

8. Nit: -b is listed first in the man page OPTIONS, after -i in usage(), and last in the getopt string -- let's pick one order.

The commit split, the tests (nice use of the hungfs factory), the man page, bash-completion and the libfuse-dev CI addition all look good.

— assisted by Claude Code